Hi, Anybody knows, if that HiRes issue was fixed in available FW updates? 02.04? Thanks.
The "flattop" screenshot definitely looks weird. At first glance it might look like "a" windowed FFT. But look closer:
- It is nothing like a flattop window. A flattop produces a wide main-lobe and the side-lobes shouldn't have a smooth envelope.
- Why do we see wide side-lobes at all? Has the number of points suddenly dropped? The RBW display hasn't changed but unfortunately that is bugged anyway (it seems to be off by a constant factor of 1000 though; that would suggest we are still at 1 Mpts, the maximum according to the datasheet).
- Look at the main-lobe again. It actually hasn't changed much at all, it is still very sharp. And so are the harmonics. So we must still be at full resolution.
Maybe there is just a typo in the formula for the window function.
It is definitely possible to construct a window that behaves like this (for example take a "normal" window function and superimpose it with a low-amplitude, much narrower window), but you wouldn't really do that on purpose.
But it does look weird...
It looks typical of most FFT images on the web.
eg. https://www.gaussianwaves.com/2020/09/window-function-figure-of-merits/
As you say, it's probably related to number of bins/number of sample points in the FFT.
No it does not.
If you look at images on internet (wikipedia for instance), they demonstrate windowing with 11 or 13 sample points..
That is why you see something similar... With 1 million points FFT you don't see it.
And if I had to choose what it looks like it looks more like Parzen window with 11 sample points......
Because of all that i presume there is some "magic" happening with Rigol's FFT choice of how many bins it uses automatically...
Also one thing to note: They state RBW : 4.999m
?? What is that
FFT is definitely buggy and to me not finished. Few critical control variables user cannot control, and are not reported.
Looks like FFT is still a toy function in any scope. It's not for measurements, it's just for reference. The values for SNR or dynamic range are not provided. The only related parameter in the spec is channel isolation, which is of 1:100. That's 40 dB, so anything on the FFT chart that is below that level (-60 dB, -100 dB, whatever) is just FYI. Accuracy is not officially specified. The most fierce debates are usually about toys
Looks like FFT is still a toy function in any scope. It's not for measurements, it's just for reference. The values for SNR or dynamic range are not provided. The only related parameter in the spec is channel isolation, which is of 1:100. That's 40 dB, so anything on the FFT chart that is below that level (-60 dB, -100 dB, whatever) is just FYI. Accuracy is not officially specified. The most fierce debates are usually about toys
That's the point, if you need reliable spectrum analysis you buy a full specified spectrum analyzer, a DSO is designed to be used mainly as YT instrument, for sure the FFT implementation is not the key point when you look for a new instrument.
What make me really laugh is that in this forum we can see dozens of DSO's FFT screen with 100dB (or even more) vertical dynamic range and infinite debates spent to demonstrate that they are actual
Anyhow, let's all keep in mind that the HDO1074 is available in EU for a little less than 1200 euro VAT included, it offers most of HDO4K features with a respectable 200MHz BW, 100Mpts , 2GSa/s.
This is the real bargain from this new line and there is nothing on the market that come close to it, i.e. the cheapest SDS2000X-HD cost 3 times more.
It looks to me like the FFT mode doesn't do real decimation after all (at least in the screenshots). The noise graphs published earlier in this thread seem to indicate that the Rigol has a bit less noise in the kHz region anyway. And it looks like it is running 1 Mpts FFT vs 512 kpts, so that is another 3 dB of processing gain. If there was real 500 MSa/s -> 5 MSa/s decimation (and not just subsampling) that should give another 20 dB and I'm not seeing that.
So I think it is still not quite clear what is happening with respect to resampling. Does the noise floor decrease if you switch acquisition mode to Hi-Res?
Btw I disagree that spectrum analysis does not matter much on an oscilloscope. The main point of having 12 bits is that you can analyze small details on large signals that you often can't see in a time-domain display. Spectrum analysis is a major part of the toolbox. I do agree that specifications lack a lot of detail compared to high-resolution options from Lecroy or R&S. But those are exactly the gaps that people are trying to fill here.
And it looks like it is running 1 Mpts FFT
Indeed, basically looks like 1MPts (~5Hz bin spacing), OTOH the displayed RBW=5mHz @5MSa/sa would imply 1Gpts, which can't be true.
However, the "flattop" trace rather suggests a window size of only ~50k points (~100Hz bin spacing), zero-padded to a larger FFT size, leading to interpolation in the frequency domain then (which makes the shape of the side lobes visible). And as you already said, the main lobe is way to narrow for a flattop window. Any common window function has a main lobe width >= 2x the width of side lobes. Something is weird here.
Looks like FFT is still a toy function in any scope. It's not for measurements, it's just for reference. The values for SNR or dynamic range are not provided. The only related parameter in the spec is channel isolation, which is of 1:100. That's 40 dB, so anything on the FFT chart that is below that level (-60 dB, -100 dB, whatever) is just FYI. Accuracy is not officially specified. The most fierce debates are usually about toys
I think that is a bit simplified look at things..
SDS2000X HD: CH to CH
Isolation (@50Ω)
> 60 dBc, < 500 MHz
> 70 dBc, < 350 MHz
Also CH2CH isolation is relevant only if you use several channels at the same time.. With single channel not so much..
At that time only SFDR is relevant......
So it is very useful but need to be realistic what it is and what is it not...
What make me really laugh is that in this forum we can see dozens of DSO's FFT screen with 100dB (or even more) vertical dynamic range and infinite debates spent to demonstrate that they are actual
A 12 bit scope cannot have 120 dB dynamic range.. But it can have -120dBm noise floor, and it can have larger than ADC dynamic range in FFT because of software gain.
Sometimes large dynamic range is not needed, good sensitivity is enough...
GPS signals are in the noise of preamp. It is literally not visible in time domain...
Using clever software techniques they extract useful signal from the noise...
Math don't lie. But it can be misapplied and misunderstood.
And I agree that it is better to be cautious and have a healthy dose of scepticism than believe in fairy tales..
But sometimes we can squeeze a bit more performance with a bit of cleverness.
Strange things gentlemen...
Playing with the FFT, same Setup as before.
Setting Memory to 100 Mpoints, hanning window.
Playing with the timebase..
From say 500ms to 50ms the memory remains the same.
Then switching to 20ms - the memory drops down to 10Mpoints.
Switching back to 50ms, the memory remains 10 and the FFT is freaking out....
You can go in the menu and set the memory to 100 again, FFT becomes normal again.
Will upload a video in a few minutes.
Looks like the FFT input is being truncated (without proper windowing). Hard to tell what exactly is going on. Some buffer size mismatch maybe?
Here is small example (dynamic). (SDS2504X HD)
And note, For this particular purpose these FFT etc settings are not at all best optimal ones.
Signals: (note that these said levels are not accurate, just entered set values using SDG1032X (but I know these levels do not have high error)
If everything is ideal and accurate these are levels.
(marker 1) 5000Hz, +22dBm
(marker 2) 5100Hz -48dBm (carrier level, AM modulation with 20Hz, and modulation depth 20%). 70dB below marker 1
(marker 3) 5120Hz -68dBm (AM USB -20dBc because 20% mod depth)
90dB below marker 1
FFT mode is just Normal and acquisition just normal, no any averaging or other things. Just normal acquisition.
FFT window Flattop.
As can see displayed levels are not bad when also think what is signal source and what are all specifications.
Hi,
the flattop thing from yesterday...
Here a video, look at the FFT on various timebases, especially how the form of the fundamental wave looks like.
https://youtu.be/83K0Y8Bov2kWeird: With the flattop window memorypoints won´t change...
The rest of the FFT test...
Playing from 500ms timebase till 5ms, watch the info on the window (RBW, samplerate), I´ve changed nothing else except the timebase.
I can not guarantee the correct order of the pics as the forum software somehow shuffled them since a couple of months.
More FFT test I don´t make, imho it´s too buggy.
Next play is connecting the bodnar pulser.
Last for today, the bodnar pulser.
I´ve been looking for switching the display mode (dot/vectors), to my surprise you can´t change the mode, it´s always vector.
(BTW, to the owners: I haven´t found the possibility to save all the screen content to the stick, when doing a screenshot, all open menus won´t be captured, so I must use a camera)
Then connecting the pulser, measure the rise time: 1.6ns sounds plausible to me, while the signal looks a little bit "filtered".
Following an idea, I´ve activated the second channel and to my surprise, the pulse on ch1 is getting "faster" and overshoots came up.
(1.2ns when 2 channels are active).
When all channels are active, the overshoot of the pulse gets "blurried" while the risetimes decreases to 1.5ns .
Hm...
Last for today, the bodnar pulser.
I´ve been looking for switching the display mode (dot/vectors), to my surprise you can´t change the mode, it´s always vector.
Purposely done to foil rf-loop's investigations of false dots.
(BTW, to the owners: I haven´t found the possibility to save all the screen content to the stick, when doing a screenshot, all open menus won´t be captured, so I must use a camera)
Inspect the Save/Recall menu for the switch to show or hide menus in screenshots.
Do you forget your Siglent has that ?
...............
Hm...
Indeed.
Inspect the Save/Recall menu for the switch to show or hide menus in screenshots.
Do you forget your Siglent has that ?
No, why should I...This was the first section I´ve looked for.
Edit:
BTW, the
symbol in the channelbox stands for 50 Ohm input...
Purposely done to foil rf-loop's investigations of false dots.
Snip of the manual:
19.1 Display Type
This series oscilloscope provides the "Vector" display mode in which the sample
points are connected by lines and displayed. In most cases, this mode can provide the
most vivid waveform for you to view the steep edge of the waveform (such as square
waveform).
LOL....
And the manual telling me another thing, about the screenshot...
There is no possibility to save the whole screen content....
(BTW, to the owners: I haven´t found the possibility to save all the screen content to the stick, when doing a screenshot, all open menus won´t be captured, so I must use a camera)
On the web interface you have menu "Print screen" . There you have both options Screenshot and Screen Capture.
Thank you but it should be also possible without web control..Something for the wishlist.
Not for me,but for the current and future owners.
Following an idea, I´ve activated the second channel and to my surprise, the pulse on ch1 is getting "faster" and overshoots came up.
(1.2ns when 2 channels are active).
When all channels are active, the overshoot of the pulse gets "blurried" while the risetimes decreases to 1.5ns .
Hm...
Sinc interpolation is supposed to introduce pre-shoot and overshoot, when applied to an undersampled edge. I guess the frontend is too fast to prevent aliasing at 1GSa/s and 2GSa/s. Is it possible to turn off sinc interpolation?
Just powered on again*, there is no possibility to do it.
*(When plug in the scope to mains, it will switch on for a second (fan starts running, display lit)before it reaches the standby condition)